The student prepared a diploma thesis focused on the prediction of atrial fibrillation (AF) recurrence after radiofrequency ablation.

The theoretical part of the thesis summarizes the current state of the art, including AF manifestation, risk factors, and available treatment options. The practical part describes the dataset as well as the entire process of model development, including training, fine-tuning, hyperparameter optimization, validation, and testing of a deep learning model for AF recurrence prediction. The work also includes an effort to address model interpretability.

The results are presented clearly and are followed by a comparison with existing methods in the Discussion chapter. The thesis has a total length of 57 pages and is supported by figures and tables. From a formal point of view, the work is at very high level, including the used language quality. Nevertheless, I would appreciate a higher amount of visual content (e.g., tables comparing methods or configurations).

The student worked with relevant literature and cited 64 sources, including books, journal articles, and conference papers. The student also worked
with current EU and US clinical guidlines, which is important for clinical context.

From a technical perspective, I appreciate the student’s approach to the problem. Since the target dataset (FNKV Hospital, Prague, Czechia) was too small for training a deep learning model from scratch, the student used transfer learning – pretraining on the CinC/PhysioNet Challenge 2021 dataset followed by fine-tuning on the FNKV dataset.

Given the computational demands of such experiments, the student utilized external computing resources (MetaCentrum computing grid), which represents an additional technical challenge successfully addressed. In this context, I would appreciate a more detailed analysis (e.g., a table or figure) illustrating how architectural modifications or hyperparameter choices influenced model performance.

In terms of applicability, I expect that the presented results may extend the findings of the FNKV research team, which is primarily focused on clinical and hand-engineered features. The presented work therefore has clear potential for future collaborative publication.

I consider the thesis to have fulfilled its objectives and recommend it for defense. I evaluate this work with 94 points (grade A). Výsledný počet bodů navržený vedoucím: 94

In his thesis entitled “Design and Implementation of a Method Estimating the Effectiveness of Radiofrequency Ablation in Patients with Atrial Fibrillation”, the author Jan Hýl developed and tested a method from the realm of deep learning to estimate the effectiveness of radiofrequency ablation in patients with Atrial Fibrillation (AF). In
particular, he implemented an approach based on transfer learning to adapt a pretrained ResNet ECG classification model to predict the risk of AF relapse. The thesis spans 47 pages (introduction to conclusion) and contains 9 figures.
The level of presentation of the thesis is excellent, with a very detailed introduction into the medical / physiological background of the topic. The level of technical detail is appropriate. The structure of the thesis is adequate, making it easy to follow and understand.
The formal editing and the language level of the thesis are also very good, with only a few grammatical errors. In particular, the figures of the thesis are well-designed and support the content. To improve matters even further, the author could have included the graphics in vector format. However, since the resolution of the figures is appropriately high, this is merely a minor point.
The thesis contains 64 references and Mr. Hýl showed comprehensive and precise work with the literature. A minor critique point is the occasional use of block citations, which could’ve been avoided by assigning references directly to the statement they support.
Overall, the level of professionalism and usability of the results is very good. In my opinion, this master’s thesis would be publishable as a conference paper with minor changes. Hence, I would rate the thesis of Mr. Jan Hýl with 90 out of 100 points. Výsledný počet bodů navržený oponentem: 90